Abstract. Using HNN extensions of the Boone-Britton group, a group E is obtained which simulates Turing machine computation in linear space and cubic time. Space in E is measured by the length of words, and time by the number of substitutions of defining relators and conjugations by generators required to convert one word to another. The space bound is used to derive a PSPACE-complete problem for a topological model of computation previously used to characterize NP-completeness and RE-completeness.Introduction. The ability of mathematical systems to simulate computation has often been used to prove unsolvability results. The first, and most instructive, example was Post's simulation of Turing machines by finitely presented semigroups [10]. For each deterministic Turing machine M, Post constructs a semigroup T(M) on generators we shall call qa, sb, where a and A range over certain finite sets. An instantaneous description (ID for short) of the state of computation at any moment can be written as a word 2 in these generators, there is a special qa generator called q, and the defining relations of T(M) are such that 2 = q is derivable if and only if the ID 2 leads M to halt. Thus the halting problem for M is reduced to the word problem for T(M), and by choosing an M with unsolvable halting problem Post proved the unsolvability of the word problem for semigroups.The simulation of M by T(M) shows that "semigroups can compute". What is interesting from the viewpoint of computational complexity is that the derivation in T(M) which reflects a given computation of M has length bounded by a linear function of the length of computation. Thus "time" in T(M) (measured by the